Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!iuvax!noose.ecn.purdue.edu!venus.ecn.purdue.edu!giacomet
From: giacomet@venus.ecn.purdue.edu (Frederic B Giacometti)
Newsgroups: comp.lang.eiffel
Subject: trouble with variant assertions
Message-ID: <1990Sep29.054436.16582@ecn.purdue.edu>
Date: 29 Sep 90 05:44:36 GMT
Sender: news@ecn.purdue.edu (USENET news)
Organization: Purdue University Engineering Computer Network
Lines: 55

I have the 2.2B version on a SunSparc. Has this feature ever been
tested before its release ? In some cases, whatever be the integer
variable or constant I put, my own trace indicates that an exception
for invariant violation is generated when intering the second loop.

Does this means that the variant is not initialized to its value and
remains zero ?

How can this happen ?

My feature is:

unit: like Current is
	local i: INTEGER
    do
	Result.Create( dim );
	from i:= 1
	 variant c_1: 100  -- you can put `dim' or 100000 the result is the same
	 until dim < i
	 loop
	   Result.set_cell( field_unit, i, i );
	   io.putint(i); io.new_line;
	   i := i+1
	end
    end;

I get:
--------------------------------------------------------------------------------
Object Class        Routine              Nature of exception Effect
--------------------------------------------------------------------------------
1ED764 MATRIX3      unit                 "c_1":
		  (From MATRIX_SQ)     Loop variant violated.  Fail  
--------------------------------------------------------------------------------

when entering the second loop (i=2)

It's the most trivial iteration. It works perfectly if I remove the
variant. Although similar, some loops work, other don't. I ended up
removing all my variants to see that everything was OK.

I don't want to be nasty; so far I've been much impressed by
most capabilities of the eiffel compiler; but does ISE test its
releases uniformly on all machines ?
I remember, for the silliest one, encountering a "rabs" fonction from the
library which would turn its non-zero negative arguments into zero's.

Practically, could ISE post a complete list of the bug reports
they have received for each platform for the 2.2 version ?
It could save us further troubles and spare us unexpected discoveries.

Thanks,

Frederic Giacometti
School of Industrial Engineering
Purdue University